Chief Operating Officer / China
Decisions that cost planned savings and what they bought
- Total assets of the merged bank
- RMB 214bn
- Xuantang Rural Commercial Bank, December 2025; not my operating budget
- Personal customers served
- 5.1 million
- Through 318 outlets and 2,140 village service points, June 2026
- Median release of a wrongly restricted account
- 3 days
- Working days, 2025; nineteen calendar days in early 2024
Decisions & consequences
01Closing thirty-eight outlets instead of sixty-one after the merger
When the eleven institutions merged in November 2023, Xuantang inherited 356 outlets. In forty-one towns two predecessors had outlets within two kilometres of each other, and twenty-three township outlets were the only bank counter in town and lost money every year. The merger plan, prepared by our chief financial officer, Wenjie Deng, proposed closing sixty-one outlets for an annual saving of RMB 74 million. Our customers include 1.4 million people over sixty, many of whom collect pensions and farm subsidies in cash and do not use mobile banking.
I proposed that no closure paper reach the board unless the outlet passed three tests: a receiving outlet within two kilometres, a pension-day queue at that outlet below forty-five minutes in a timed trial, and a bus journey of under forty minutes for nine in ten customers over sixty-five. Thirty-eight outlets passed. I recommended keeping twelve loss-making single outlets and moving eleven to three days a week. Wenjie argued, in a memo that went to the board with my paper, that the merger case depended on the full saving and that loss-making outlets would only become harder to close. Our president, Hanqing Zhou, supported my proposal on condition that the twelve were reviewed every year.
We gave up about RMB 31 million of annual savings. Staff from each closed outlet moved with their customers to the receiving outlet for six months. We bought nine mobile service vehicles and wrote to every account holder ninety days before a closure.
By June 2026 we had closed the thirty-eight. Deposits held by the closed outlets' customers were 96% of their pre-closure level after a year. Complaints about closures fell from 214 in the first year to 61 in the second, and the median pension-day queue at receiving outlets was twenty-six minutes. Every network paper now includes the three tests and a timed pension-day trial.
The calendar is what I would change. We closed a mountain-town outlet in January 2025, two weeks before the Spring Festival, and on the first pension day the queue at the receiving outlet passed two hours. We added a temporary desk within a week, but older customers stood in the cold to teach us that. I would also have met the village committees before drawing the closure list, not after.
Three processing sites instead of one
Each of the eleven predecessor institutions ran its own back office: about 1,020 people in all, with different voucher forms, authorisation limits and cut-off times. At the larger ones a counter waited a median 7 minutes 40 seconds for authorisation of a large or unusual transaction, and 1.8% of processed vouchers came back with errors. An external consulting firm proposed a single centre in Xuantang city with 520 staff, saving RMB 98 million a year.
I argued for three sites, in Xuantang city, Dongpu and a southern county seat, with 610 staff and every process able to run at two of them. My reason was an outage I had lived through in 2012 at Tongze Commercial Bank, when a single centre was cut off for five hours and counters across nine branches stalled. Our chief financial officer and the consultants pointed out that my design saved RMB 72 million rather than RMB 98 million and kept costs in counties where we had too many back-office staff. Our president asked for a tabletop exercise of a site loss before deciding, and after it he supported three sites.
We gave up about RMB 26 million a year. We migrated process by process rather than county by county, cut 340 local voucher forms to 96 and published counter waiting times weekly. Of the 410 colleagues whose back-office work ended, 46 joined fraud review, 140 the new credit operations unit, 130 the customer service centre and 94 the counters. Nobody was made redundant, and each move came with paid retraining.
By the end of 2025 the median authorisation wait was 1 minute 50 seconds and the voucher error rate 0.5%. In July 2025 flooding closed the southern site for three days, and its work moved to Dongpu within two hours without a counter queue anyone noticed. The risk management committee now receives an annual failover test.
What I would change is the order of training. We switched counters in two counties before their staff had practised on the new forms, and errors there stayed above 1% for four months. I also underestimated how many customers over seventy preferred handwritten slips; we restored a paper option for them in the second quarter.
A three-day clock for accounts our own rules had restricted
In early 2024 Xuantang had 23,600 personal accounts under restrictions created by its own fraud-monitoring rules, mostly suspensions of mobile and non-counter payments. The rules came from eleven predecessors and overlapped. When we sampled 600 restricted accounts, 71% belonged to customers with an ordinary explanation: migrant workers paid by several employers, older people receiving transfers from their children, traders collecting from many buyers. A customer had to return to the outlet where the account was opened, sometimes from a city a thousand kilometres away, and the median wait for a decision was nineteen days. Restrictions were our largest complaint category, with 3,900 complaints in 2023.
The stakes ran both ways. The anti-telecom and online fraud law requires banks to manage risky accounts, and an account we released could be used to move stolen money. Our head of legal, compliance and anti-fraud, Xiaodong Hu, opposed any release standard, arguing that the bank's exposure to penalties was asymmetric. I proposed that every restriction created by our own rules carry an owner and a three-working-day review clock, that any outlet could verify the customer and that video verification be available for customers far away. Restrictions instructed by the police were outside the proposal.
We agreed a compromise. Accounts receiving funds from several unrelated sources would get a second review, and a quarterly sample of released accounts would go to the risk management committee. We gave up RMB 21 million a year for a 46-person review team, and some new restrictions were handled more slowly while the team was trained.
By mid-2025 the median release time was three working days. Restriction complaints fell to 1,150 in 2025. Of released accounts, 0.4% were later found to have been misused, a figure the committee reviews every quarter. After we retired fourteen of thirty-seven overlapping rules, the number of restricted accounts fell to 9,800. Customer-reported telecom fraud losses fell from RMB 38.6 million in 2023 to RMB 14.2 million in 2025, mostly because of counter interventions with older customers.
I would change the order of the work. We should have retired the duplicate rules before building the team, which would have spared customers several months. Our first explanation letter was also written in legal language that older customers could not follow; we rewrote it with six of them.
Taking the paperwork out of small-business and farm lending
In 2024 Xuantang had RMB 46 billion of inclusive loans to small businesses and farm households, about 186,000 borrowers and 1,150 loan officers spread across the counties. A median small-business or farm loan took twenty-one working days from application to disbursement. When my team traced 400 files, about four of those days were spent on the credit judgement itself. The rest went on collecting missing documents, re-keying data, routing approvals and registering collateral. Post-loan reviews found documentation defects in 31% of files, and loan officers told us they spent close to half their week on paperwork. Large banks were lending in our counties at lower rates, and farmers who needed money before spring planting did not wait three weeks.
I proposed a central credit operations unit to take document checking, contract preparation, collateral registration and disbursement away from the counties, while leaving the credit decision with the county loan officer and the chief risk officer's approval chain. Our head of inclusive finance, Kaiwen Lu, wanted to go further and approve loans of up to RMB 300,000 centrally and automatically from tax and utility data. I opposed that for first-time and thin-file borrowers, whose risk the data could not see. Our chief risk officer, Ruolan Tang, sided with keeping the judgement local. We settled on automatic renewal only for repeat borrowers with two clean years. Three county branch presidents objected to losing staff and control over files.
We gave up the speed full automation promised and some county autonomy. The unit of 140 people opened in May 2024, staffed largely from the merged back offices.
By the end of 2025 median turnaround was eight working days and the defect rate 9%. Loan officers' paperwork time fell to about a quarter of their week, and inclusive loans grew to RMB 52 billion. The non-performing ratio on 2024 and 2025 new inclusive loans was 1.6% against 2.4% for the pre-merger book, though those loans are young and I do not read much into the comparison yet.
What I would change is the order. We started with contracts because they were easiest; collateral registration was the longest wait, and we should have begun there. We also staffed for an average month and met the March planting peak with a backlog of nine days.